-- QiongHu - 12 Dec 2012

  1. Result using normal selected point instead of the harmonic point is better, so finding the original paper about why they use discrete cepstrum and LPC is necessary. (And the difference) (Sending papers about this)
  2. As the un-voiced frame has been solved, next is to synthesize speech with sinusodial reconstructed spectrum together with parameter like phase and jitter using STRAIGHT.
  3. Print 3D figure under frequency, time and spectrum amplitude SpectrumlVideo.tiff[1].
  4. Next meeting is 10am with toshiba
  5. Proposal: adding references even for draft like the following[2]

  6. Draw the normal cepstrum for comparing (middle of the envelope or not)
  7. Solve the problem of shifting
  8. For selecting all the FFT point not the harmonic point, the envelope is not accurate and easy to be affected.
  9. Comparing the envelope between reconstruted spectrum using all FFT point and harmonic point and original Straight.
  10. For drawing figures, like seldom point, order should be changed.
  11. Select accurate harmonic peak (paper, hand, or artificial way)
  12. Shiga thesis


[1]sent by Junichi


Topic attachments
I Attachment Action Size Date Who Comment
pngtiff SpectrumlVideo.tiff manage 280.8 K 13 Dec 2012 - 10:20 QiongHu  
Topic revision: r3 - 13 Dec 2012 - 15:04:45 - QiongHu
This site is powered by the TWiki collaboration platformCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback
This Wiki uses Cookies